Functional Correctness on CPP-Bench
38.1Functional Success Rate @1IG-CD
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| IG-CDBackbone Model=DiffuC-7B, Decoding Method=IG-CD2026.01 | 38.1 | 38.3 | 38.3 | 38.5 | |
| LAVEBackbone Model=DiffuC-7B, Decoding Method=LAVE2026.01 | 38.1 | 38.4 | 38.5 | 38.5 | |
| NO-CDBackbone Model=DiffuC-7B, Decoding Method=NO-CD2026.01 | 37.2 | 37.4 | 37.8 | 37.8 | |
| FS-CDBackbone Model=DiffuC-7B, Decoding Method=FS-CD2026.01 | 35.2 | 35.5 | 35.8 | 35.9 | |
| LAVEBackbone Model=Dream-7B, Decoding Method=LAVE2026.01 | 33.5 | 38.2 | 39.5 | 41.4 | |
| IG-CDBackbone Model=Dream-7B, Decoding Method=IG-CD2026.01 | 29.1 | 30.5 | 30.8 | 31.1 | |
| FS-CDBackbone Model=Dream-7B, Decoding Method=FS-CD2026.01 | 26.5 | 26.8 | 27.1 | 27.5 | |
| NO-CDBackbone Model=Dream-7B, Decoding Method=NO-CD2026.01 | 25.6 | 26.2 | 26.8 | 28.1 | |
| LAVEBackbone Model=LLaDA-8B, Decoding Method=LAVE2026.01 | 19.5 | 23.8 | 25.9 | 26.7 | |
| LAVEBackbone Model=LLaDA-1.5, Decoding Method=LAVE2026.01 | 19.1 | 23 | 24.9 | 28.8 | |
| IG-CDBackbone Model=LLaDA-1.5, Decoding Method=IG-CD2026.01 | 18.3 | 21.3 | 23.1 | 27.1 | |
| IG-CDBackbone Model=LLaDA-8B, Decoding Method=IG-CD2026.01 | 17.8 | 22.5 | 23.7 | 25.6 | |
| NO-CDBackbone Model=LLaDA-1.5, Decoding Method=NO-CD2026.01 | 17.2 | 20.1 | 22 | 25.9 | |
| NO-CDBackbone Model=LLaDA-8B, Decoding Method=NO-CD2026.01 | 16.3 | 21.8 | 23.1 | 24.9 | |
| FS-CDBackbone Model=LLaDA-1.5, Decoding Method=FS-CD2026.01 | 15.9 | 16.1 | 16.5 | 16.5 | |
| FS-CDBackbone Model=LLaDA-8B, Decoding Method=FS-CD2026.01 | 15.7 | 16 | 16 | 16.2 |